Traditionally, seismic processing back-propagates a surface recorded wavefield to its subsurface reflectors and thereby creates a wiggly image of the subsurface. The back-propagated wavefield enables an interpreter to derive a model of the structural and stratigraphic geology of the imaged subsurface (, , ). The interpreter integrates the initial structural image with additional geological and geophysical information to a complete geological subsurface model.
In the last decade, geophysical researchers have attempted to extract information beyond the wiggly, structural image from high-quality seismic recordings. Wavefield inversion Bleistein (1984); Tarantola (1987); Weglein (1989) estimates physical rock properties, such as local P-, S-wave velocity, and density. Amplitude-versus-Offset analysis relates the prestack seismic data to pore fluid contents or hydrocarbon indicators (). Seismic attribute mapping empirically correlates pre- or poststack seismic data characteristics, e.g., instantaneous frequency, to rock properties Sonneland et al. (1990); Taner et al. (1979). Many studies combine the various approaches - inversion, AVO, seismic attributes - with other data - e.g., well log data - to extract improved lithological and petrophysical subsurface information (). The overall goal of this research is to generate continuous, reliable, geological models of the subsurface.
While much work has concentrated on widening the application of seismic data to extract additional information, limited research has been done to improve the extraction of the traditional structural interpretation. Today's interpreters explore huge three-dimensional image volumes by displaying them on graphics workstations (); Sonneland (1983) and by plotting traditional paper sections. These media force interpreters to identify three-dimensional features - river channels, faults - on two-dimensional cross-sections of the image volume. Because of the wavefield character of the seismic image, the interpreter lacks a depth view of the features of interest. To improve picking accuracy, interpreters may use seismic event picking software that evaluates a picked image location by local volume correlation.
Amoco's fault detection technique, Coherency Cube Bahorich and Farmer (1995), aims to help interpreters by transforming a traditional seismic image volume to a discontinuity image volume. Marfurt et al. published an improved scheme. In principle, both scheme's resemble the plane-wave misfit, I introduce in this paper. However, I replaced Marfurt's compute-intensive search for the plane-wave dip by a simple analytic estimation. Coherency Cube generated much interest and, subsequently, other researchers (); Gersztenkorn and Marfurt (1996) presented similar techniques. Bednar presented a discontinuity attribute based on least-squares dip estimation. Today many commercial seismic processing and interpretation packages include a discontinuity attribute process; their methodologies are usually confidential or patented.
Fundamentally, image enhancement amplifies the pixel amplitudes of the target features relative to the amplitude of the undesired background features. Transform methods, such as Fourier, Wavelet, Hough, or Principal Component decomposition, explicitly separate the image into components that can then be recombined using target enhancing weights. Prediction methods, on the other hand, assume the absence of the target feature and produce high amplitude output for image pixels that deviate from that assumption. However, transform and prediction methods are two modes of thinking rather than two distinct approaches: the Laplace operator, a standard edge enhancement technique (); Jain (1989), can be thought of as a prediction of the central pixel amplitude derived from its surrounding pixel values. Alternatively, the Laplace operator is often implemented by Fourier transform, in which case it is best thought of as amplifying the high-frequency image components.